A cost benefit analysis of management alternatives for first trimester miscarriage: Results from a discrete choice experiment within a randomised controlled trial Emma McIntosh1 and Stavros Petrou2 1

Corresponding author Senior Research Officer Health Economics Research Centre Department of Public Health University of Oxford Old Road Campus Headington Oxford, OX3 7LF Tel: 01865 226634 Fax: 01865 226842 [email protected] 2

Health economist National Perinatal Epidemiology Unit University of Oxford Old Road Campus Headington Oxford, OX3 7LF Tel: 01865 226829 Fax: 01865 227002 [email protected]

Paper presented at the Health Economists Study Group meeting, Glasgow, 30th June to 2nd July 2004 First Draft

Work in Progress: Please do not quote, cite or copy without permission of the authors

Acknowledgements: The authors are grateful to Professor Mandy Ryan for designing the discrete choice experiment.

Background Approximately one in seven confirmed early pregnancies end in miscarriage during the first trimester (Regan, Braude, & Trembath 1989). The traditional management approach for miscarriage is surgical evacuation of the retained products of conception. This approach was developed in the first half of the 20th century as a result of high rates of gynaecological infection from the retained products of conception and the ensuing mortality from septicemia (Ankum, Wieringa-de Waard , & Bindels 2001). Expectant management or ‘watchful waiting’ does not involve surgical or medical intervention and is another alternative in the management of first trimester miscarriage. Women choosing expectant management do not require ward stay, surgery or medical treatment and are advised to rest at home. Expectant management has recently been acknowledged by many general practitioners as a means of facilitating a well regulated natural process in human reproduction. Finally, medical management of first trimester miscarriage involves the use of drugs (Mifepristone, Misoprostol) to expel the retained products of conception. This management option often involves hospital stay, usually outpatient admittance. A number of randomized controlled trials (RCTs) comparing expectant management with surgical management (Chipchase & James 1997;Nielsen & Hahlin 1995; Wieringa-de Waard

et al. 2002), medical management with surgical management (Chung et al.

1999;Creinin, Moyer, & Guido 1997;De Jonge et al. 1995;Demetroulis et al. 2001;Johnson et al. 1997) and expectant management with medical management (Nielsen, Hahlin, & Platz-Christensen 1999) have been conducted. However, no RCT to date has compared all three management methods. Moreover, the RCTs that have been conducted lack information on resource use, thus preventing assessments of costeffectiveness from being made. In addition to this, no trials to date have explored, withintrial women’s preferences for attributes of all three management options using economic methodology. This paper will focus on the use of the economic methodology of discrete choice experiments (DCEs) to elicit womens preferences for attributes of alternative forms of miscarriage management. One of the uses of DCE data is to estimate willingness to pay values for use in cost-benefit analyses (Kleinman L et al. 2002;Lancsar & Savage 2004;McIntosh, Donaldson, & Ryan 1999;Ratcliffe 2000;Ryan 2004). Prior to reporting

2

the methods and results of the study, this paper will take a brief look at previous studies in this area in addition to describing the clinical trial alongside which this miscarriage DCE was carried out. Exploring women’s preferences for using discrete choice methodology A small number of papers have used the economic methodology of DCEs in the area of maternal and perinatal health (Hundley, Ryan, & Graham 2001; Longworth, Ratcliffe, & Boulton 2001;Petrou & Henderson 2003;Ryan M & Hughes J 1997). Ryan and Hughes (1997) used the methodology of DCEs to explore women’s preferences for attributes of surgical and medical management of miscarriage: the DCE survey was sent to women following a randomized clinical trial of surgical and medical management of miscarriage. In the study by Ryan and Hughes (1997), the authors estimated willingness to pay as well as utility score estimates for attributes of miscarriage. Attributes included the following: level of pain experienced; time in hospital receiving treatment; time taken to return to normal household activities; cost to women of treatment; and complications following treatment. The design of this DCE was to compare all scenarios with the ‘current situation’ of surgical management. The results showed that all attributes were significant predictors of choice of management and a negative constant in the probit model implied there was a general preference for surgical over medical management. The MIST Trial The DCE reported in this paper represented part of the economic research conducted alongside a randomized controlled trial (RCT) of alternative management methods of first trimester miscarriage (The MIST Trial) (Trinder et al. 2004). The economic evaluation in the MIST trial provides for the first time, both the clinical information on all three alternative miscarriage management options (expectant, medical and surgical) and information on the use of resources thereby allowing the cost-effectiveness of expectant, medical and surgical management of first trimester miscarriage to be assessed (Petrou et al. 2004). Furthermore, the MIST study was also novel in that it was the first RCT of miscarriage management to incorporate information on both resource use and willingness to pay (WTP) estimates (derived via a DCE questionnaire to women in the

3

trial). The incorporation of this element allows a cost-benefit analysis (CBA) to be reported along with the cost-effectiveness analysis (CEA) results. This paper will concentrate upon the CBA element of the MIST trial. The paper represents work in progress and will concentrate upon a number of issues specific to the use of DCE methodology. These include: ♦

The use of DCEs within RCTs

♦

The appropriate choice of levels for post-analysis welfare estimation for policy purposes

♦

The incorporation of choice probabilities into the welfare estimates

♦

The use of DCE-derived willingness to pay estimates within health care costbenefit analyses

♦

The use of cost-benefit analyses more generally for allocating health care resources

Methods Women with a confirmed pregnancy of less than 13 weeks gestation, who had miscarried, were randomized into the three arms of the MIST trial outlined above: expectant, medical or surgical management. Women allocated to the expectant group were allowed home with no intervention. The management of women allocated to the medical group depended on the type of miscarriage. Those with incomplete miscarriages were admitted to hospital and given a single vaginal dose of 800µg misoprostol, whilst those with missed miscarriages were pre-treated with a single oral dose of 200mg mifepristone and then admitted 24-48 hours later for a single vaginal dose of 800µg misoprostol. Women allocated to surgical management were admitted for surgical evacuation of the retained products of conception in line with the usual policy of six participating clinical centres. Documented gynaecological infection within 14 days of trial entry constituted the primary clinical outcome of the trial. This was defined as two or more of the following: purulent vaginal discharge; pyrexia >38.0 C; tenderness over the uterus on abdominal examination and/or increase in white cell count (WCC) above 15x109/ml. A prospective economic evaluation was conducted alongside the MIST trial, which is discussed in detail

4

in the paper by Petrou et al. (2004). In brief, the use of health services and other resources by women up to 8 weeks post-miscarriage was collected using a combination of trial data collection forms, observational research and self-completed questionnaires and then combined with unit costs (£, 2001-2 prices) to obtain a net societal cost per woman. The nonparametric bootstrap method was used to present cost-effectiveness acceptability curves and net benefit statistics at alternative willingness to pay thresholds held by decision makers for preventing one gynaecological infection. A DCE was run in parallel to the MIST economic evaluation. Women participating in the RCT received a DCE postal questionnaire three months following their miscarriage which elicited their preferences for the two management alternatives that they did not receive. For example, women in the surgical management arm received a DCE questionnaire which elicited preferences for medical versus expectant management. The aim of such an approach was to control for the effect of women expressing a preference for the actual management option they had received. Following on from the work carried out by Ryan and Hughes (1997), the attributes and levels for characteristics of miscarriage management included in this DCE are shown in Table 1 below.

5

Table 1

Attributes and levels for MIST DCE

Attribute

Levels

Time spent at the hospital receiving treatment

Overnight Half a day One day

Level of pain experienced

Low Moderate Severe

Number of days of bleeding following treatment

3 Days 8 Days 14 Days

Time taken to return to normal activities after 1-2 Days treatment

3-4 Days 7 days or more

Cost to women of treatment

£50 £150 £250

Chance of complications requiring more time or Very unlikely (about 5 in 100) readmission to hospital

Quite unlikely (about 10 in 100) Unlikely (about 20 in 100)

The attributes and levels were combined into scenarios using statistical software and set alongside the ‘current situation’ scenario of surgical, medical or expectant management. Hence, three different questionnaires were designed such that women in each arm of the trial were asked their preferences for the two management alternatives not received. Questionnaires 1, 2 and 3 were therefore administered to women who had undergone expectant, medical and surgical management respectively. Since a constant ‘current situation’ increases the length of the questionnaire (as that option remains fixed throughout), each questionnaire contained 25 choice sets. Twenty five choices was felt to be too lengthy and hence each questionnaire was blocked into 2 questionnaires (1a, 1b, 2a, 2b, 3a & 3b), with either 12 or 13 choices. For analysis, however, the individual ‘a’

6

and ‘b’ blocks for designs 1, 2 and 3 were combined and analysed as one questionnaire. Each questionnaire contained between 2 - 4 naturally occurring consistency checks. The data were analysed using a random effects probit model to account for the repeated observations nature of the data and a constant term was included to allow for testing for inherent preferences for the labeled options (surgery, medical or expectant) when assuming the attributes and levels are equivalent. Willingness to pay values were obtained by estimating the marginal rate of substitution (MRS) between the attribute coefficients and the cost coefficient. Confidence intervals around the WTP values were obtained using the variance-covariance matrix of the coefficients and the cost coefficient (Propper 1990). Incremental WTP and incremental costs were then placed together within a formal CBA framework using scenarios from the pre-trial questionnaire and again using scenarios obtained from the actual trial data (post-trial scenarios). Net benefit values and cost/benefit ratios were calculated for shifts between the management options.

Results Questionnaire 1 was administered to 198 patients who underwent expectant management. Once missing ‘prefer’ variables were removed, the analysis was carried out on 189 patients (2,331 observations). Questionnaire 2 was administered to 228 patients who underwent medical management. Once missing ‘prefer’ variables were removed, the analysis was carried out on 223 patients (2,771 observations). Questionnaire 3 was administered to 222 patients who underwent surgical management. Once missing ‘prefer’ variables were removed, the analysis was carried out on 218 patients (2,711 observations). The overall consistency rate for all 6 questionnaires was 83.6%. For details of the individual consistency test results please see Appendix 1. The results from the random effects probit analyses for all three questionnaires are presented in Tables 2 - 4 below.

7

Table 2

Values for expectant managed women – surgical versus medical options (surgical treatment fixed comparator)

Variable Constant Time Bleeding Activities Cost Complications Pain Level (ref=low) - Moderate - Severe

WTP (£) per unit increment/ decrement (95% CI’s)

Attribute Unit

Coefficients

SE

P

/ Days Days Days £ %

0.2991 -0.043644 -0.05496 -0.16602 -0.003177 -0.05866

0.1692 0.0056 0.0103 0.0210 0.0006 0.0092

0.08